Message-ID: <4745881B.6010904@archonet.com>
Date: Thu, 22 Nov 2007 13:46:03 +0000
From: Richard Huxton <dev@archonet.com>
User-Agent: Thunderbird 2.0.0.9 (X11/20071031)
MIME-Version: 1.0
To: Richard Broersma Jr <rabroersma@yahoo.com>
CC: SQL Postgresql List <pgsql-sql@postgresql.org>,
	General PostgreSQL List <pgsql-general@postgresql.org>
Subject: Re: [SQL] Bad Schema Design or Useful Trick?
References: <82640.215.qm@web31815.mail.mud.yahoo.com>
In-Reply-To: <82640.215.qm@web31815.mail.mud.yahoo.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Richard Broersma Jr wrote:
> Below I've included sample table definitions for a vertically
> partitioned disjunctive table hierarchy.  I wanted to point out the
> use of the composite primary key declaration that is applied to two
> columns that are clearly not a candidate key.  However, using the
> badly defined primary key allows for referential integrity to nicely
> handle the de-normalization between the main table and sub tables
> that is inherent with this type of data model.
> 
> Would using a primary key in this manner be a decision that I will
> regret in the long run? If so, can any explain why?
> 
> The parent table is parts with the child table pumps and hardware.
> 
> CREATE TABLE Parts ( part_nbr varchar( 100 ) UNIQUE NOT NULL, 
> part_type varchar( 20 )  NOT NULL,
...
> PRIMARY KEY    ( part_nbr, part_type ),
...

So - what you're saying is that because part_type depends on part_nbr it 
shouldn't be part of the key, but because you want to search by 
part-type in the referencing tables it makes life easier.

Will you regret this? Probably - I always seem to end regretting making 
short-cuts, although in this case I can't see any direct harm that could 
occur.

I'd probably make (part_nbr) the pkey and have a separate unique 
constraint on (part_nbr,part_type) that I reference. That "feels" better 
, although I'm not sure it actually gains you anything.

-- 
   Richard Huxton
   Archonet Ltd